Under-mining health? Local impacts of iron ore mining in India

 

Abstract

Environmental justice concerns about adverse impacts of extractive mining industries

challenge their claim of broad-based development. However, empirical evidence of negative

externalities attributable to mining enterprises is thin. In this paper, we examine the burden of

Acute Respiratory Illness and Malaria among villagers along a gradient of proximity to mining

areas to generate evidence pertinent to the debate on environmental health and social justice. By

focusing on health outcomes as human development indicators, the study also reflects on

whether the ‘resource curse’ that has been posited at the country level also has micro footprints.

The analysis combines 600 household interviews conducted in 20 villages with GIS data

on locations of mines and villages in India. Self-reported health data on incidence and workdays

lost due to Acute Respiratory Illness and Malaria are outcomes of interest. Estimation results for

incidence (Probit) and workdays lost (3SLS and Zero-Inflated Negative Binomial) indicate that

villagers living closer to mines reported higher frequency of ARI complications and workdays

lost due to Malaria. However, villagers living closer to mines are more likely to be employed in

mines, and this in turn positively affects adoption of improved stoves that can potentially reduce

indoor air pollution.

(198 words)

I. Introduction

Concerns about environmental justice challenge the premise of broad-based development

promised by extractive mining industries in developing countries. In India, vehement political

opposition from grassroots socio-environmental activists and public interest groups highlight the

poor track record of state-supported large-scale extractive projects in conservation of the local

environment and improving welfare of the affected populations (Sarangi, 2004; Randeira, 2003;

Guha, 1993). Alienation from the political mainstream (Rew, 2006), state-ownership of mines

(Sinha et al, 2007) and an inadequate approach to participatory development deprives the local

population an appropriate share of the rent on the resource (Campbell, 2003). Weak enforcement

mechanisms of environmental standards fail to keep pollution within limits and subsequent

adverse effects on the local population (Blacksmith Institute, 2008; Weber-Fahr, 2002). Critique

of development policy from the perspective of environmental justice focuses on this disparity in

the distribution of benefits and costs from mining. In this paper, we provide empirical evidence

relevant to this debate and pertinent to impact assessment of development policy across the

world.

The situation also resembles a micro footprint of the ‘Resource Curse’ hypothesis

proposed by Sachs and Warner (1997, 2001) at the national level – the consistent negative

correlation between economic performance (per capita GDP growth rate) and natural resource

abundance (share of exports of natural resource-based products). Bulte et al. (2005) have

extended the literature on the resource curse by considering a broader set of welfare and

development criteria; they find that natural resource abundance is negatively associated with

indicators of nutrition, poverty and life expectancy. By focusing on the human population that is

both positioned to take advantage of new economic opportunities and most exposed to the

environmental effects arising from mining, we can assess the specific pathways and micro-

distribution of mining impacts. This level of analysis contributes to understanding the reasons

for the resource curse and the possibilities for mitigating that curse through appropriate benefit-

sharing mechanisms (Hancock, 2002).

The state of Orissa in India faces this challenge as it embarks upon a major reform

program with the mining sector taking center stage in the growth process. Much of the proposed

mining expansion is in remote regions with a predominantly tribal population. Previous instances

of expansion of iron ore mines in the state were marred by expropriation of tribal land and large-

scale displacement of villagers1. In this paper, we empirically examine the consequences of

mineral extraction on health outcomes of people living close to mines. We (1) contribute to the

renewed interest in linking public health with use of environmental resources (Lee, 2002), (2)

provide empirical evidence for the ‘resource curse’ at the micro-level and (3) examine if

employment in mines influences adoption of prevention measures (like improved cook stoves

and bed nets). From a survey of the literature on health issues arising from mining activities,

acute respiratory illness (ARI) (Stephens and Ahern, 2001; Joyce, 1998) and malaria (Webster,

2001) were found to be most commonly cited besides occupational injuries. However, the

literature on impact of iron ore mining on community health, especially in the context of

developing countries is thin. Using cross-sectional data, we develop multivariate econometric

models to distinguish between environmental and occupational health pathways through which

mines could have an impact on health outcomes of individuals. We combine information from

household surveys with spatial information on location of mines and villages to investigate

associations between health outcomes, employment and proximity to mines. While we observe

no strong associations for occupational health outcomes using information on employment in

mines, we find that ARI incidences increase among individuals living closer to mines, but

malaria incidence in the population increase in villages further away from mines. Analyzing if

employment in mines eases financial constraints for households to adopt preventive measures

(e.g. improved stoves and mosquito nets to reduce the incidence of ARI and malaria

respectively), we find that it positively influences adoption of improved stoves.

II. Local health consequences of mining

Freudenberg and Wilson (2002) in a review of case studies on local socio-economic

impacts of mining in United States challenge the belief that mining leads to rural development.

Their results, on the contrary, point to higher observed levels of poverty and unemployment in

the mining areas. Mining projects around the world have come under severe criticism under

counts of land expropriation and environmental degradation that harm the livelihoods and health

                                                       1 New York Times, January 13 , 2006  

of local communities (Keenan et al 2002; Sosa 2000). Mining projects involve huge investments

accompanied with strong political influence, and local communities could bear substantial

environmental, economic, and social costs unless local governments enforce strong regulatory

systems to ensure equitable sharing of benefits (Auty, 2006). An independent assessment of

World Bank sponsored mining projects in India concluded that ‘people living close to mines

have suffered most and usually benefitted least’ (Teri, 2001). However, there are few studies

focusing on the health impacts of mining from a policy perspective (Hilson, 2002).

A central tenet of research on the environmental impacts of mining has been that because

mines occupy a relatively small land area (when compared, for example, to other land uses like

forestry or agriculture), the effects of mining on the environment will be localized (Bridge,

2004). Though dispersal of toxic wastes and other harmful byproducts of mining by wind and

water happen over wide geographical areas, the detrimental impacts are most pronounced in

areas close to mines. There are direct and indirect pathways that mines affect health outcomes of

people. Employment in mines is an example of a direct pathway for impacts of mines, both in

terms of the benefits from employment in mines as well as occupational health outcomes. Mines

could also affect welfare though environmental health pathways that is direct as well as indirect.

For example, declining ambient air quality due to spread of dust and chemicals from mining

areas directly affect people living close by, irrespective of whether they work in mines or not

(Sinha et al., 2007, Stephens and Ahern, 2001). On the other hand, deforestation due to mining

activities indirectly affects families dependent on forest products for income generation and

nutritional requirements by reducing their access to the resource base (Peters et al. 1989).

Among public health concerns for mine workers, incidence of respiratory disorders has

received considerable attention in the environmental and occupational health literature (Ross and

Murray, 2004; ILO, 1997)2. Mining in both surface and underground mines involves drilling and

shearing of large quantities of minerals. The clouds of dust raised in displacing these materials

can severely damage the lungs, particularly after years of exposure (Joyce, 1998). Occupational

exposure to air pollutants has been found to be a major cause for chronic cough and asthma,

symptoms common in chronic bronchitis (Hedlund et al., 2006). However, there is a negative

externality to society as individuals not employed but living close to mining areas are also

                                                       2 For a review of the occupational and community health impacts of mining internationally, refer to Stephens and Ahern, 2001.

exposed to the harmful effects of air pollution. Prevalence of acute and chronic respiratory health

has been observed among individuals close to open cast or open pit mines (Pless Mulloli et al.,

2000). In a study on opencast coal mining in the state of Orissa in India, suspended particulate

matter (SPM) were found to be significantly higher than permissible limits in the mines as well

as in the surrounding locations from various operations (Chaulya, 2004). In other studies on open

cast coal mines, poorly maintained dirt roads and movement of heavy vehicles to transport the

ore resulted in dispersion of coal dust in areas adjoining mines and cause severe air pollution

(Ghosh and Majee, 2000; Singh and Sharma, 1992).

Incidence of malaria is also common in areas close to mining activities, though the

pathways of health impacts are both direct and indirect. Borrow pits left after road constructions,

drains and abandoned excavation areas in opencast mines often increases breeding sites for the

malaria vector and directly increases malaria prevalence (Yasuoka and Levins, 2007).

Deforestation caused by mining activities and subsequent change in land use and human

settlement alters the local ecosystem, changes the vector ecology of mosquitos and indirectly

affects malaria incidence (Patz et al., 2004, Takken at al. 2003). Among all states in India, the

incidence of malaria is the highest in the state of Orissa (Kumar et al., 2007). A cross-sectional

study conducted in 1989 in settlements in the iron ore mining region in Orissa found high

densities of the malaria vectors. Children were found to be most vulnerable to malaria attacks

and the poor casual laborers in the mines were found to be worst affected in economic terms

(Yadav et al., 1991).

III. Prevention options to mitigate incidence of ARI and malaria

We focus on adoption of improved cook stoves and bed nets by households that should

also influence the incidence of ARI and malaria respectively. Our rationale to focus on these

interventions are three-fold – (1) failure to account for such preventive measures will confound

the association we aim to establish between mining and incidence of ARI and malaria; (2) a

positive association between employment in mines and adoption of such preventive measures

will offset the adverse health impacts of mining on the local population; (3) examine if micro-

credit and social learning provide practical solutions to financial and information constraints that

negatively influence the adoption of these measures.

A recent simulation study by Wilkinson et al. (2007) provides evidence of significant

reduction in mortality due to respiratory complications from adoption of improved cook stoves

and reduction in indoor air pollution (IAP). The products of incomplete combustion from

biomass combustion (firewood, crop residue) contain a number of health-damaging pollutants

(Fullerton et al., 2007; Bruce et al., 2000) responsible for a significant global burden of

respiratory infections (Smith et al., 2004; WHO)3. 70% of the population living in rural India

(and majority in our study) traditionally cooks using unvented stoves with women and children

experiencing the highest exposures to indoor air pollution. As India embarks on a ‘National

Biomass Cookstoves Initiative’4 with the policy objectives of social and health development at

the forefront, it is relevant to examine factors that influence the adoption of these stoves. In a

recent study of indoor air pollution in Bangladesh, construction of the kitchen with proper

ventilation was found to yield better indoor air quality (Dasgupta et al., 2007).

As compared to other methods of malaria vector control like indoor residual spraying of

DDT, use of bed nets is technologically simpler and cost-effective (Misra, 1999; Lengeler,

1996). In a randomized trial of insecticide-treated bed nets in Sundargarh district of Orissa

(adjacent to location of the present study), relative risk of malaria and parasite rates declined

significantly in villages with treated nets (Sharma et al., 2006). Partnerships between donor

agencies, local NGO and communities have been established to raise the awareness and

disseminate insecticide-treated bed nets (Barat, 2006).

In spite of the claimed benefits of these preventive measures in reducing the incidence of

ARI and malaria, the rate of adoption has been consistently below expectations of the agencies

and NGOs promoting them. Much of the inertia in adoption is attributed to financial constraints

facing rural families (Dasgupta et al., 2007; Wallmo, 1998 for Cookstoves; Meltzer, 2003;

Nuwaha, 2001 for bednets). These empirical findings suggest that increased cash income – for

example, from employment in mines – could increase adoption rates of these preventions. Also,

positive effects of collective learning through information sharing among neighbors has been

observed in de-worming among children (Miguel and Kremer, 2004) and improved sanitation

practices among rural families (Dickinson and Pattanayak, 2009).

                                                       3 http://www.who.int/mediacentre/factsheets/fs292/en/index.html - downloaded April 25, 2008 4 http://mnes.nic.in/press-releases/press-release-02122009.pdf - downloaded January 7, 2010 

IV. Study area

The state of Orissa lies along the eastern coast of India with the largest reserve of

superior quality hematite iron ore in the country (Sengupta, 2005). Situated along the Northern

border of the state, Keonjhar district was selected for this study because of the concentration of

iron ore mines in Joda block within the district5. 31% of the total mining employment in the state

of Orissa is concentrated in the district of Keonjhar indicating the importance of the mining

industry in the region. Mining for iron ore in the district began in the 1950s, and much of the

planned expansion and liberalization of the mining sector in Orissa will open up new mining

areas in this region. The recorded forest area in Orissa in 2003 was 4.84 million hectares, which

constituted 31.06% of the geographic area and ranked fourth among Indian states in terms of

total forest cover6. However, in comparison with 1999, forest cover had decreased by almost a

million hectares7. The district had a relatively high percentage (42.7%) of forest cover in 1999

(Forest Survey of India, 1999). But, in the two blocks selected for this study, analysis of the

classified land cover data reveals that 13.4 square kilometers of vegetative cover were replaced

by expanding mining areas between 1989 and 2004.

In the absence of field monitoring and measurement data on ambient air quality, study

villages were selected based on proximity to mines. The first stage of the sample selection

process followed a quasi-experimental design, to test the hypothesis that villagers living closer to

mines were more likely to be impacted by the mines than those further away. On the basis of

government census data on mine employment and location of mines, two blocks were selected in

Keonjhar district – Joda and Keonjhar Sadar (Figure 1). Joda block has a large concentration of

iron ore mines, as confirmed by the fact that 68% of mine workers in Keonjhar district live in

Joda block. On the contrary, Keonjhar Sadar block has a much lower concentration of mines and

only 1% of people employed in the mining industry live in this block. In the second sampling

stage in each of these blocks, 10 villages were selected at random (Figure 2). Finally, in each of

                                                       5 Keonjhar district has 20% of all the mining leases granted by the State government of Orissa. The district has more than 12076 hectares of iron ore mining areas under 46 mining leases, making it the most important iron ore mining center in Eastern India (Source: Status paper on mining leases in India. Vasundhara, India). 6 ‘State of Forest Report’ published b y the Forest Survey of India in 2003; http://www.fsiorg.net/fsi2003/states/index.asp?state_code=21&state_name=Orissa 7 ‘State of Forest Report’ published by the Forest Survey of India in 1999; http://www.envfor.nic.in/fsi/sfr99/sfr.html 

these villages, 30 households were selected from each village to be interviewed. For

convenience, villagers in the Joda block henceforth will be referred to living in the ‘high

exposure’ zone while Keonjhar Sadar will be referred to as the ‘low exposure’ zone.

V. Description of the data

The dataset consists of two elements – (1) a household survey administered to 600

households with specific modules on questions on incidence of ARI and malaria among family

members; (2) forest cover maps derived from Landsat satellite imagery for 1989 and 2004, and

GIS information on locations of mines and villages. For indicators of health outcomes, we use

reported information on types and days of illness specifically pertaining to incidence and

workdays lost due to of ARI and malaria8. Information on the number of iron ore mines was

collected from the database maintained by the directorate of mines. Information from survey

topo-sheets and remotely sensed data was combined with the mining database to identify the

location and area of iron ore mines in the study area. These mines are under state as well as

private ownership, operating under various levels of modernization and mechanization.

Observations from field visits to a sample of mines in the study area revealed apparent variations

in abidance to environmental regulations prescribed by the Department of Mines, which has

repercussions on occupational and environmental health issues. For exposure to mines, GIS

information on location of mines and villages are used to construct a proxy measure based on

Euclidean distance to the nearest iron ore mine9. As a result of the study design, villages in the

high exposure and low exposure zones are within a range of 0.2 to 4 km and 6 to 21 km from

iron ore mines respectively. Note that all the villages fall within the “Peripheral Development

Zone” of 50 km10 - a specification followed in mining and development projects supported by

                                                       8 Unless mentioned, incidence and workdays lost due to ARI is based on the reported information on whether an individual had suffered from ARI in the year prior to the survey in 2005. For malaria, this is based on information from 2001 to 2005. 9 Alternative definitions of exposure based on ‘number of mines in a 2km buffer around each village’ were used in the analyses which provided similar results. WE stick to using the Euclidean distance based measure for exposure to mines in this paper. Besides the mining areas, there are unregistered, small-scale stone-crushing units in the study area and the dust from these units cause significant air pollution. However, these units do not operate in the same spot for a long period of time to avoid impoundment, and their effect on ARI specifically is difficult to assess. Based on data and field observations, proximity of villages to these stone-crusher units is highly correlated to distance to iron ore mines. 10 World Bank report on the study ‘Towards Sustainable Mineral Intensive Growth in Orissa, India’ is available at:

the World Bank to address issues related to local impacts of mines. The descriptive statistics of

the variables used in the analysis is reported in Table 1.

A total of 600 households (300 in each block) were interviewed and information was

collected for 2949 individuals. Out of 600 households, 175 out of 300 households interviewed in

the high exposure zone reported at least one family member to be working in the mines, while

the corresponding number in the low exposure villages was 45 out of 300 households11. 283

individuals from 216 families reported to be working in mines12. Table 2 shows the distribution

of health outcome variables in the full household sample and the two blocks separately. The

average number of reported cases of ARI in the family was significantly higher in the high

exposure zones as compared to the low exposure zones (1.2 compared to 0.8). However, the

average number of reported cases of malaria in the family was higher in the low exposure zone

(1.8 compared to 1.5). While the average number of workdays lost per family due to ARI was

not significantly different across exposure zones, families in the high exposure zones reported

more number of workdays lost due to malaria (19.5 compared to 12.5). A reason for this trend

could be the fact that incidences of malaria are lower in the high exposure zone, but the severity

of malaria is worse in the high exposure zone. Correlations with distance to iron ore mines show

that workdays lost increases among villages with greater proximity to mines. Note that the mean

and variance of count of ARI and malaria variables are similar and thus we model these two

variables as a Poisson process in our econometric estimation.

Table 3 presents the results on health status breaking down the population into 3 sub-

groups – (i) individuals with no family members working in mines, (ii) individuals with some

family member working in mines, (iii) individuals working in mines themselves. Parsing the

sample into these subgroups helps to better distinguish the occupational health impact on

individuals from mine employment. While individuals in group (i) should have no occupational

health effect, those in group (ii) only have an indirect effect as members working in mines could

                                                                                                                                                                               http://www_wds.worldbank.org/external/default/WDSContentServer/WDSP/IB/2007/12/21/000020953_20071221103718/Rendered/PDF/398780IN.pdf 11 The household questionnaire had modules to record the occupation details of every household member. Employment of a family members in mines was determined for the cases where a member was reported to be working as a ‘non-farm worker as factory worker’. Employment in the stone quarries was determined based on member occupation being ‘non-farm work as construction labor’. 12 Another 103 individuals from 65 families reported to be working in stone quarries. As mentioned before, these stone quarries operate illegally and employment in these units is temporary in nature. WE ignore quarry employment in the subsequent analyses.

transmit ARI or malaria contracted while working in mines among other family members. Group

(iii) represents the sub-sample most likely to suffer a direct occupational health effect from

working in mines. There is no significant difference in ARI related health indicators across these

three groups, except for expenditure where individuals not working in mines report having spent

more. For malaria, those who work in mines suffer more in terms of workdays lost as well as

significantly spend more on treatment.

VI. Conceptual framework

Figure 3 presents a conceptual framework to analyze the possible impact of mining on

public health. Mining operations can generate both direct and indirect health impacts for the

proximate population, and these pathways are categorized as environmental and occupational.

There are direct financial benefits from employment in a mine if the opportunity cost of working

in the mines is lower than the mining wage. The increase in income can reduce the cash

constraints of families to invest more in illness prevention measures that improve overall health

status. Proximity to mines has direct and indirect negative health effects as well. Mines that fail

to meet prescribed environmental standards are more likely to generate direct negative

externalities for those who are employed in mines (occupational health), as well as through

negative externalities for those who do not work in mines but live close to it (environmental

health). In the context of the paper, the dust and other harmful suspended particulate matter

could increase ARI-related health problems, or abandoned areas with stagnant water become

breeding grounds for mosquitos that cause malaria. Large areas had to be deforested to establish

the open cast mines in the region. This imposed a financial and nutritional burden on the

villagers in the study area who relied on sale and consumption of forest resources for their

livelihood.

This framework provides a more nuanced description of the complex set of relationships

that connect mining with public health concerns. It helps separate causal pathways, provides a set

of testable hypotheses and the empirical analyses required to verify those (Ethridge, 2004). The

specific hypotheses that we focus on in the econometric section are: (a) proximity to mines

increase incidence of ARI and Malaria; (b) income from mines reduces financial constraints for

farmers and enables them to invest in preventive activities that reduces the burden of illness.

VII. Econometric models for health outcomes

In the first step of the econometric model, the individual-level dataset is used to model

two separate outcomes related to ARI and malaria – (i) whether an individual member reported

the incidence of ARI and malaria; (ii) the reported number of workdays lost by individual

members due to ARI and malaria. The empirical model for both cases is outlined as follows.

Let ( )y be the outcome of interest – either incidence or workdays lost due to ARI and

malaria. These health outcomes are a function of a vector of individual-specific

characteristics ( )I , household-specific characteristics ( )H , location of the household ( )L ,

environmental health variables ( )EH , occupational health variables ( )OH and preventive

behavior ( )P .

( )POHEHLHIFy ,,,,,= (1)

where ( ).F depends on whether incidence of ARI/Malaria (a binary variable) or workdays lost

due to ARI/Malaria (count variable) is being modeled.

The variables used in each group are:

Individual-specific ( )I : Age, sex, education, number of hours devoted to wage employment on an

average day by individual member.

Household-specific ( )H : Caste, amount of fuelwood used for domestic consumption, amount of

irrigated land in village.

Location of household ( )L : distance to paved roads, distance to primary health center.

Environmental health ( )EH : distance to nearest iron ore mine using GIS information.

Occupational health ( )OH : dummy if individual is employed in mines, number of days reported

to be working in mines.

Preventive behavior ( )P : Use of improves stoves for cooking, dummy if kitchen is partitioned

from the house or not, dummy if bed nets are used for sleeping.

(a) Probit model for incidence of ARI and malaria:

A binary variable was constructed based on the information of which individual family

member had occurrence of ARI and malaria. A Probit of the following form is estimated:

( ) ( )XGXy ββ +== 01Pr , where a vector of explanatory variables ( )X is specified in (1), and

( ).G is the standard normal cumulative density function. Individuals within the family have the

choice of working in mines, which makes mine employment endogenous to the model. WE use

proximity to mines, average number of hours worked in wage employment, dummy if land

owned and years of education as instruments for mine employment and estimate the Probit as a

2SLS model.

Results from the Probit model are shown in Table 4. The environmental health variable,

measured by Euclidean distance to the closest iron ore mine is significant across the two models,

but with opposite signs. While as distance to mines increase by 1 km, the probability of a family

member being affected by ARI problems decrease by 2.3%, while the probability of malaria

increase by 1.9%. This result mirrors the previous observation that villagers closer to mines in

the high exposure zone reported a higher incidence of ARI, while those further from mines in the

low exposure zone reported a higher occurrence of malaria. Employment in mines, after being

instrumented, is negatively affected with the incidence of ARI – an additional day of

employment in mines reduces the probability of ARI by 0.1%. Employment in mines increases

the probability of malaria, though the result is not significant. None of the demographic

variables, except whether the family belongs to a scheduled tribe, is significant. Generally,

people belonging to the scheduled tribes are historically marginalized, and 70% of the mine

workers in our sample belong to scheduled tribes. This result raises the doubt of under-reporting

of health status information by this group. Location-specific variables like distance to paved

roads and primary health center had little explanatory power. We use the log of amount of

fuelwood consumed by the household to control for the fact that combustion of firewood for

domestic cooking and heating could also lead to higher incidence of ARI. If this factor is

ignored, then the impact of mines on ARI incidence could be over-estimated. Increase in

firewood use has the expected effect of increasing ARI, but the effect is not significant. In the

case of malaria, We control for the amount of irrigated land and change in forest cover around

the village – factors that could affect the incidence of malaria. The significantly negative effect

of irrigated land on incidence of malaria is surprising, as Patz (2004) argues that irrigated

agricultural fields often provide breeding grounds for mosquitos. Given the clustered nature of

these villages, agricultural land is arranged in concentric circles around the villages. Irrigation in

agricultural fields often results in stagnant pools of water in the land surrounding the villages and

was suspected to provide breeding areas for mosquitos. However, given the highly seasonal

nature of rain-fed irrigation and the economic importance of agricultural land for households, the

irrigation variable might be working as a proxy for welfare in the low exposure villages. Villages

with a greater decrease in forest cover in a 2km buffer around the village reported higher

incidence of malaria, but the effect was not significant. The prevention variables all have

expected signs and are highly significant – incidence of ARI reduces by 75% and 23% among

individuals who have improved stoves and partitioned kitchens in their house respectively.

Incidence of malaria reduces among individuals by almost 12% if they use insecticide-treated

bed nets in the house.

(b) 3SLS models for workdays lost due to ARI and malaria:

In the sample, workdays lost due to ARI and malaria has a range of 0 to 90 days each. In

this model, workday lost is treated as a continuous variable and a 3SLS model is developed.

Employment in mine is assumed to be endogenous and is estimated simultaneously with

workdays lost due to ARI and malaria. The error term in both equations is assumed to be

correlated due to unobserved heterogeneity and 3SLS estimation method is employed.

Table 5 shows the results from these two models. The environmental health variable

remains significant and negative for both the ARI and malaria models. Recall that Table 2

indicated that though the incidence of malaria was high in the low exposure zones, the number of

workdays lost due to malaria was higher in the high exposure zone leading to the conjecture that

the severity of malaria is worse in villages closer to the mines. These general observations

substantiate the negative coefficient with distance to iron ore mines. Living 1km closer to the

iron ore mines increases the workdays lost due to ARI and malaria by 0.02 and 0.08 days. The

instrumented occupation health variable is not significant across either of the models. Older

individuals suffer more days of lost work due to ARI. Individuals belonging to the scheduled

tribes and scheduled castes report less workdays lost due to both ARI and malaria. Compared to

models in table 3, increase in forest cover in a 2km buffer around villages (comparing 1989 and

2004 classified land cover data) is found to increase the workdays lost due to malaria. Loss in

forest cover in a 2km buffer around sample villages was higher in the low exposure zones than in

the high exposure zones (2.3 sq. km compared to 0.31 sq km), and the spatial distribution of

greater malaria incidence in the low exposure zones supports the deforestation-malaria

hypothesis. However, malaria incidence or workdays lost cannot be attributed to deforestation

due to mining activities. Location-specific variables still remain largely insignificant, except that

workdays lost due to ARI are higher for individuals living further away from the primary health

center. The prevention variables continue to have the expected signs, but only kitchen

construction and use of bed nets are significant.

(c) Count data models for workdays lost due to ARI and Malaria:

While the range for reported workdays lost by individuals due to ARI and Malaria ranged

between 0 and 90 days, more than 99% of the sample lost between 0 and 20 days for each

illness13. Unlike treating workdays lost as a continuous variable as in the 3SLS models estimated

in (ii), an event count model is estimated. As noted in Table 2, the difference between mean and

variance of workdays lost due to ARI and malaria with the variance being higher indicated

overdispersion in the data (Long and Freese, 2006). There are also a large proportion of

individuals who did not report workdays lost due to illness. We assume two different processes

generating the data – one explaining the zeros, and the other for the positive number of workdays

lost (Lambert, 1992). We thus use Zero-Inflated Negative Binomial (ZINB) models to account

for overdispersion and excess zero that we observe in the data (Cameron and Trivedi, 1998). A

battery of visual and model fit statistics are compared to determine that the ZINB model was

most appropriate compared to ordinary Poisson or Negative Binomial models.

The results from the ZINB model are presented in Table 6. Comparing results with the

3SLS results in Table 5, the occupational health variable (days worked in mines) is insignificant

and becomes negative in the zero-inflated models. The environmental health variable (distance to

nearest iron ore mine) is consistently negative but insignificant14. The prevention variables all

have expected signs, but only use of improved stoves remains significant in explaining reduction

in workdays lost due to ARI. Amount of firewood consumed is surprisingly significantly

negative across all models, another result that deviates from the 3SLS model. Male family                                                        13 For ARI (malaria), out of 631 (1009) individuals reporting workdays lost to illness, only 14 (5) reported to have lost more than 20 days due to ARI. These models are thus run with the restricted sample where reported workdays lost due to ARI and malaria was less than or equal to 20 days. 14 When the logarithm of distance to nearest to iron ore mines is used instead, the coefficient is just significant at the 15% level. 

members reported to have lost more workdays lost due to ARI, which is likely as most of the

wage labor are done by male members and are more likely to report loss of working days

compared to female members mostly engaged in household chores. An individual living 1 km

closer to an iron ore mine will increase the number of workdays lost to ARI by 0.3%. Being a

male reduces workdays lost by 13%, raising the concern that women members in the household

are more susceptible to ARI related problems, though the use of firewood for cooking is a critical

factor for that. There is no perceptible difference in workdays lost for individuals with mine

employment. Families with improved stoves and partitioned kitchen reduce workdays lost to

ARI by 11% and 2% respectively. In the model for malaria in Table 6, the occupational health

variable remains positive but insignificant across all models. The environmental health variable

is significant and negative, indicating that while incidence of malaria is lower in villages closer

to mines, those living closer report to have suffered more from workdays lost due to malaria. The

use of mosquito nets reduce workdays lost due to malaria but the result is insignificant.

Workdays lost increases with more land in irrigation and greater loss of forest cover around the

village, as was surmised from the literature review. An individual living 1 km closer to an iron

ore mine will increase the number of workdays lost to malaria by 7%. A 1% increase in area

under irrigation around the village increases workdays lost to malaria by 3%, while 1 sq. km. of

deforestation around the village increases workdays lost by 1.5%. A family with an additional

mosquito net reduces workdays lost to malaria for individual members by 11%.

As a summary of the results from these sets of model, there is little evidence of

occupational health for either ARI or malaria. The environmental health effect, using the proxy

of proximity to mines, is more pronounced, as evident for ARI in the 3SLS model and on malaria

in the ZINB model. The hypothesis that higher deforestation leads to higher prevalence of

malaria was only supported in the ZINB model, though this deforestation could not be attributed

to mining activities alone as more deforestation took place in the low exposure zone. Another

robust result from the models reflects the importance of prevention activities on individual health

outcomes. Preventive measures like adoption of improved stoves and construction of partitioned

kitchen inside homes reduced the incidence and workdays lost due to ARI. Use of insecticide-

treated bed nets also reduced the incidence and workdays lost due to malaria. From a public

health policy perspective, it is important to identify factors that influence the adoption of these

prevention measures.

VII. Factors affecting use of prevention measures

Families make decisions regarding the adoption of the three preventive measures based

on their subjective utility post-adoption. It is important to point out that these prevention

measures help reduce the burden of ARI and malaria related health problems among family

members, irrespective of whether these problems were induced by mining activities or not. The

typical approach to model adoption behavior of individuals is to discrete choice model. Among

the factors that could affect adoption, most interest lies in the fact that whether cash income from

employment in mines eases the cash constraint facing poor, rural households thus facilitating

investment in preventive measures. There are also opportunities for individuals to observe the

changes in outcomes among neighbors in the village who use these prevention methods.

This kind of exchange of information among residents in the village often influences

adoption decisions of individuals (Dickinson and Pattanayak, 2009). In the model, evidence for

such kind of social interaction effect in adoption of these preventive measures is also tested in

the model. The household’s latent utility from adoption of preventive measures can be modeled

as set of household specific and location-specific characteristics, such that

( ) iii eBXbB += −1* ,

where, iB is the individual adoption decision, iB− denotes the adoption decision of other

members in the village and iX is a set of household and location-specific characteristics.

This model is estimated using a probit regression such that the likelihood function is of the form:

( ) ( )iiii BXFBXB −− −−−== θβ1,1Pr , where, ( ).F is the standard normal cumulative density

function.

The results of the estimation are presented in Table 7. Three sets of models are presented

relevant to each of the preventive behaviors. Given the economic importance of agriculture in the

region, dummy for land ownership represents wealth status of a household. Based on the dummy

for land ownership, households owning land had a 14% (24%) higher probability of using bed

nets (improved stoves). As was the case with employment in mines in previous models, days

worked by an individual in mines is treated as endogenous and instrumented using average time

devoted by family members to wage employment and Distance to iron ore mines. The

probability of adoption of improved stoves is found to significantly increase with employment in

mines. As mentioned before, dissemination of improved stoves in rural areas in Orissa are being

promoted through micro-credit organizations (Duflo, 2007). These organizations are yet to be

operational on a large scale in the region and the lack of significance of participation in micro-

credit organizations in the promotion of improved stoves or bed nets indicates this. As a matter

of fact, 72 of the households that reported having improved stoves in the region were not

members of any micro-credit groups in their village. Among the factors used to examine the

impact of social influence on household adoption decision, two variables were considered –

count of adoption by other members in the village and number of social organizations that

members of the individual households participate in. Only in case of adoption of bed nets was

the village count of adoption significant, indicating that individual adoption of bed net increases

by 3.6% when one more family in the village adopts. Participation in village level organizations

increases the scope of interaction and sharing experiences with other villagers. Such social

organizations and NGOs have been found to be important nodes of information regarding

adoption of new technology (Bandiera and Rasul, 2006). This variable was significant only in the

case of adoption of bed nets.

The cross-sectional analysis precludes inferences of any causal nature such as mine

employment causes greater use of preventive measures. However, the results point to the

possible effect where increase in cash income from mine employment allows cash-constrained

rural families to better invest in prevention of ARI and malaria.

VIII. Conclusion

This study is one of the first attempts towards comprehensive analyses of health impacts

of mining on the local population, an important stakeholder in the public policy debate

surrounding the proposed expansion and privatization of the mining industry in Orissa. We find

consistent environmental health impacts as villagers living in close proximity to mines have

higher incidences of ARI and lose more workdays due to malaria. Though the absence of

baseline data does not allow any causal inferences, the association between proximity to mining

areas and health outcomes are robust across different models that we estimate. There is evidence

that employment in mines positively influencing adoption of improved stoves that can reduce the

burden of ARI, but similar effects are not observed for adoption of bednets. Instead, there is

evidence of a demonstration effect in adoption of bednets as families use more nets if the rest of

their village also does the same.

These results resonate with the concerns raised from the environmental justice

perspective. Poor people living closer to mining areas bear a health burden, and qualitative

evidence from stakeholder interviews with government officials and local non-governmental

organizations revealed no compensation arrangements to offset such negative externalities.

These results are important from the point of distributional equity that demands a rethink about

compensation implicit in the ‘participatory development’ framework.

The analyses would have been more rigorous if clinical data was available instead of self-

reported health outcomes. Actual data from air quality monitors across the study area for the

exposure measure would have been preferred using Euclidean distances of villages from mines

as exposure. These are issues that future research on this issue needs to develop on.

References Auty, R. 2006. Mining enclave to economic catalysts: Large mineral projects in developing countries. Brown Journal of World Affairs. Vol. XIII, Issue 1, Fall/Winter. Bandiera, O., and WE. Rasul. "Social Networks and Technology Adoption in Northern Mozambique." The Economic Journal 116(2006): 869‐902. Barat, L. M. 2006. Four malaria success stories: How malaria burden was successfully reduced in Brazsil, Eritrea, India and Vietnam. American Journal of Tropical Medicine and Hygiene. 74(1), 12-16. Blacksmith Institute. 2008. The World’s Worst Pollution Problems: Top Ten of the Toxic Twenty. Downloaded from http://www.worstpolluted.org/ on December 20, 2009. Bridge, G. 2004. Contested terrain: Mining and the environment. Annual Review of Environmental Resources. 29, 205-29. Bruce, N., R. Perez-Padilla and R. Albalak, 2000: Indoor air pollution in developing countries: a major environmental and public health challenge. B. World Health Organization. 78, 1097-1092.

Bulte, E.H, R. Damania and R. Deacon (2005): Resource Intensity, Institutions and Development. World Development Volume 33 (7) 1029-1044.

Cameron, A. C. and Trivedi, P. K. 1986. Econometric models based on count data: comparison and application of some estimators. Journal of applied Econometrics. 1, 29-53. Cameron, A. C. and Trivedi, P. K. 1998. Regression Analysis of Count Data. Cambridge University Press. Campbell, M. L. and Vainio-Mattila, A. 2003. Participatory Development and Community-Based Conservation: Opportunities Missed for Lessons Learned? Human Ecology, Vol. 31 (3), 417-437. Chaulya, S. K. 2004. Spatial and temporal variations of SPM, RPM, SO2 and NOx concentrations in an opencast coal mining area. Journal of Environmental Monitoring. Vol. 6, 134-142. Dasgupta, S., Huq, M., Khaliquzzaman. M. and Wheeler, D. 2007. Improving Indoor Air Quality for Poor Families: A Controlled Experiment in Bangladesh. World Bank Policy Research Working Paper 4422. Dudka, S. and Adriano, D. C. 1997. Environmental impacts of metal ore mining and processing : A review. Journal of Environmental Quality. Vol. 26(3), 590-602.

Duflo, E., Greenstone, M. and Hanna, R. 2007. Indoor Air Pollution, Health and Economic Well-being. Working paper, Poverty Action Lab. Dickinson, K. and Pattanayak, S. K. 2009. Open sky latrines: Do social interactions influence decisions to use toilets? Working paper. Ethridge, D. E. 2004. Research methodology in applied economics: organizing, planning and conducting economic research. Iowa State University Press. Freudenburg, W. R. and Wilson, L. J. 2002. Mining the data: Analyzing the economic implications of mining for nonmetropolitan regions. Sociological Inquiry. 72(4), 549-75. Fullerton, D. G., N. Bruce and S. B. Gordon. 2007. Indoor air pollution from biomass fuel smoke is a major health concern in the developing world. Transactions of the Royal Society of Tropical Medicine and Hygiene 102, 843—851. Geist, H.J. and Lambin, E. F. 2002. Proximate causes and underlying driving forces of tropical deforestation. Bioscience, Vol 52, No 2, 143-150. Ghose, M. K. and Majee, S. R. 2000. Assessment of the impact on the air environment due to opencast coal mining: an Indian case study. Atmospheric Environment. 34, 2791-2796. Guha, R. 1993. This Fissured Land: An Ecological History of India. University of California Press.

Hancock, G. 2002. Sharing the Risks and Rewards of Mining. World Bank Mining Sector Technical Assistance Project. http://www.wmmf.org/historical/2002docs/miningpeople/Mining_People_Hancock.pdf

Hartwick, J., 1977. Intergenerational equity and investing rent from renewable resources. American Economic Review 67 (5), 972–976. Hedlund, U., Järvholm, B., Lundbäck, b. 2006. Persistence of respiratory symptoms in ex-underground iron ore miners. Occupational Medicine. 56, 380-385. Hilson, G. 2002. Small scale mining and its socio-economic impact in developing countries. Natural Resources Forum. 26, 3-13. International Labour Office. 1997. Encyclopedia of Occupational Health and Safety, 4th edn. Geneva: International Labour Office.

Joyce, S. 1998. Major Issues in Miner Health. Environmental Health Perspectives, Vol. 106, No. 11, A538-A543.

Kumar, A., Valecha, N., Jain, T., Dash, A. P. 2007. Burden of Malaria in India: Retrospective and Prospective View. The American Journal of Tropical Medicine and Hygiene. 77(6_Suppl), 69-78.

Lambert, d. 1992. Zero-inflated Poisson regression with an application to defects in manufacturing. Technometrics, 34, 1-14.

Lee, C. 2002. Environmental Justice: Building a Unified Vision of Health and the Environment. Environmental Health Perspectives. Vol. 110, Supplement 2, 141-144.

Lengeler, C. and Snow, R. W. 1996. From efficacy to effectiveness: insecticide-treated bednets in Africa. Bulletin of World Health Organization. 74 (3): 325-332.

Long, J. S. and Freese, J. 2006. Regression models for categorical dependent variables using Stata. 2nd edition. Stata Press. Meltzer, M. WE. et al. 2003. The household-level economics of using Permethrin-treated bed nets to prevent malaria in children less than five years of age. American Journal of Tropical Medicine and Hygiene. 68(4 suppl), 149-160. Miguel, E. and M. Kremer. 2004. Worms: Identifying impacts on education and health in the presence of treatment externalities. Econometrica. 72(1), 159- 217. Misra, S. P. et al. 1999. Malaria control: bed nets or spraying? Spray versus treated nets using deltamethrin – a community randomized trial in India. Transactions of the Royal Society of Tropical Medicine and Hygiene. 93, 456-457. NCAER. 1993. Evaluation survey of National Program on Improved Chulha, National Council for Applied Economic Research (NCAER), New Delhi. Nuwaha, F. 2001. Factors influencing the use of bed nets in Mbarara municipality of Uganda. American Journal of Tropical Medicine and Hygiene. 65 (6), 877-882.

Patz, J.A. et al. 2004. Unhealthy Landscapes: Policy Recommendations on Land Use Change and Infectious Disease. Environmental Health Perspectives, Vol. 112, No. 10, 1092-1098.

Peters, C. M., A. H. Gentry, and R. O. Mendelsohn. 1989. Valuation of an Amazonian rainforest. Nature 339:655–656.

Pless Mulloli, T., D. Howel, et al. 2000. Living near opencast coal mining sites and children's respiratory health.” Occupational and Environmental Medicine. 57(3): 145-51. Randeira, S. 2003. Glocalization of Law: Environmental Justice, World Bank, NGOs and the Cunning State in India. Current Sociology, 51 (3/4), 305-328. Rasmussen, R.O. and N.E. Koro leva, ed. 2003. Social and Environmental Impacts in the North . Rew, A. and S. Khan. 2006. The Moral Setting for Governance in Keonjhar: The Cultural Framing of Public Episodes and Development Processes in Northern Orissa, India. Oxford Development Studies, Volume 34, Issue 1, 99 – 115.

Ripley, E.A., R.E. Redman, and A.A. Crowder. 1996. Environmental Effects of Mining. Ross, M. H. and Murray, J. 2004. Occupational respiratory disease in mining. Occupational Medicine, 54, 304-310. Ross, Michael L. 2001. Extractive Sectors and the Poor: An Oxfam America report. Boston, MA: Oxfam America. (Available at http://www.oxfamamerica.org/art545.html) Sachs, J.D and A.M. Warner, 1997. Natural Resource Abundance and Economic Growth, NBER Working Paper Series, WP 5398, Cambridge: National Bureau of Economic Research. Sachs, J.D and A.M. Warner, 2001. The Curse of Natural Resources, European Economic Review. 45: 827-838. Sarangi, D. 2004. Mining ‘Development’ and MNC’s. Economic and Political Weekly, Vol. 39, No. 17, 1648-1652. Saxena, N.C., G. Singh, and R. Gosh. 2002. Environmental Management in Mining Areas. Sharma, S. K. et al. 2006. Effectiveness of mosquito nets treated with a tablet formulation of Deltamethrin for malaria control in a hyperendemic tribal area of Sundargarh district, Orissa, India. Journal of the American Mosquito Control Association. 22(1), 111-118.

Sinha, S., Bhattacharya, R. N., Banerjee, R. 2007. Surface iron ore mining in eastern India and local level sustainability. Resources Policy. Volume 32, Issues 1-2, 57-68.

Singh. G. and Sharma, P. K. 1992. A study of spatial distribution of air pollutants in some coal mining areas of Raniganj coalfield, India. Environment International. Volume 18, Issue 2, 191-200.

Smith, K.R., J. Zhang, R. Uma, V.V.N. Kishore and M.A.K. Khalil, 2000: Greenhouse implications of household fuels: an analysis for India. Annual Review of Energy and Environment, 25, 741-763. Stephens, C. and Ahern, M. 2001. Worker and Community Health Impacts Related to Mining Operations Internationally: A Rapid Review of the Literature. Report prepared for the Mining, Minerals and Sustainable Development project. International Institute for Environment and Development (IIED). Tata Energy Research Institute (TERI). 2001. Overview of mining and mineral industry in India. Report commissioned by the Mining, Minerals and Sustainable Development project. International Institute for Environment and Development (IIED). Takken, W., Vilarinhos, P., Schneider, P., Santos, F. dos .2003. Effects of environmental change on malaria in the Amazon region of Brazil. In Environmental change and malaria risk: global and

local implications: Proceedings of the Frontis workshop on Environmental Change and Malaria Risk: Global and Local Implications, Wageningen, The Netherlands 12-14 November; Wageningen UR Frontis series Vol. 9, 113 - 123. Wallmo, K, and Jacobson, S. K. A social and environmental evaluation of fuel-efficient cook-stoves and conservation in Uganda. Environmental Conservation 25(2): 99–108. Weber-Fahr, M. 2002. Treasure or Trouble?: Mining in Developing Countries. The World Bank and International Finance Corporation report. Webster, D. 2001. Malaria kills one child every 30 seconds. Journal of Public Health Policy. 22(1), 23-33. Wilkinson, P., K. Smith et al. 2009. Public health benefits of strategies to reduce greenhouse-gas emissions: household energy. Lancet. Vol 374 (9705), 1917-1929. World Bank Report 2007. Towards sustainable mineral–intensive growth in Orissa - managing environmental and social impacts. Environment Unit, Sustainable Development Department, South Asia Region. Report No. 39878-IN.

Yadav, R. S., Ghosh, S. K., Chand, S. K., Kumar, A. 1991. Prevalence of malaria and economic loss in two major iron ore mines in Sundargarh District, Orissa. Indian Journal of Malariology. Vol. 28(2), 105-113.

Yadav, R. S., Sampath, R. R. and Sharma, V. P. 2001. Deltamethrin trated bed nets for control of malaria transmitted by Anopheles culcifacies in India. Journal of Medical Entolmology. 38(5), 613-622. Yasuoka, J. and Levins, R. 2007. Impact of deforestation and agricultural development on Anopheline ecology and malaria epidemiology. American Journal of Tropical Medicine and Hygiene. 76(3), 450-60.    

Appendix  

 Figure 1. Two blocks chosen for the study – Joda and Keonjhar Sadar. According to the national census in 1991, 68% of reported mine workers in the district were in Joda block while Keonjhar Sadar reported only 1% of mine workers.

 

 Figure 2. Classified land cover images showing the two blocks Joda and Keonjhar Sadar. The GIS data on the location of the mines and villages were used to calculate two measures of exposure to iron ore mines – Euclidean distance to iron ore mines and number of mines in a 2km buffer around each village

        

 Table 1. Descriptive statistics of household and individual level variable used in the analyses

 

Variable Obs Mean Std. Dev. Min Max

Scheduled Caste 600 0.07 0.25 0 1

Scheduled Tribe 600 0.68 0.47 0 1

Log net firewood consumed (kg) 600 3.85 1.37 0 6.58

Participation in number of social organizations 600 1.79 1.28 0 6

Number rooms in house 600 2.70 1.48 1 11

Average amount of irrigated land around village 600 2.78 3.72 0 12

Change in forest cover within 2km village buffer (2004-1989) 600 1.24 8.43 -19.62 24.86

Log distance to paved road (km) 600 2.87 1.05 0.69 5.20

Log distance to health center (km) 600 3.41 1.06 0.69 5.08

Log distance to market center (km) 600 3.81 0.78 0.0 5.19

Distance to iron ore mine (km) 600 7.33 6.11 0.21 20.92

Cash income from mines (Rs) 600 6979.5 12383.9 0 108000

Use of improved stoves 600 0.01 0.10 0 1

Use of partitioned kitchen 600 0.53 0.50 0 1

Use of bed nets 600 0.53 0.50 0 1

Participate in micro-credit organizations 600 0.12 0.33 0 1

Number of males 2494 0.51 0.50 0 1

Age of individuals 2494 26.92 17.98 0.23 105

Education of individuals 2494 2.97 4.04 0 18

Average hours working in mines 2494 1.66 2,67 0 12

Table 2. Descriptive statistics of health outcome indicators for the full sample, and the two blocks separately  

Full sample Joda (High exposure)

Keonjhar (Low exposure)

Variable

Mean SD Min Max Mean SD Mean SD

Population weighted

mean

Correlation with

distance to iron ore

mine

Count of ARI incidence* 1.0 1.2 0 6 1.2 1.3 0.8 1.2 1.0 -0.16 †

Count of Malaria incidence* 1.6 1.4 0 10 1.5 1.2 1.8 1.6 1.6 0.12 †

Expenditure on ARI (Rs) 143.8 450.3 0 8750 158.5 287.0 129.0 568.8 146.1 -0.02

Expenditure on Malaria* (Rs) 544.4 1110.1 0 17350 678.5 1349.0 410.2 782.9 557.2 -0.09 †

Workdays lost due to ARI 4.8 9.8 0 90 5.5 9.0 4.0 10.6 5.1 -0.09 †

Workdays lost due to Malaria* 16.0 17.4 0 160 19.5 19.6 12.5 13.9 17.0 -0.16 †

* t-test indicating means of Joda and Keonjhar blocks are significantly different at 5% level

† significance at 5% level

  

Table 3. Description of health indicators by sub-groups based on mine employment  

Variable No member in family works in mine

Other member in family working in mine

Only individuals in family working in mine

mean st error mean st error mean st error

Reported ARI 0.21 0.01 0.23 0.01 0.22 0.02

Workdays lost to ARI 1.04 0.09 1.07 0.16 0.92 0.16

Expenditure on ARI (Rs.) 33.64 5.07 28.74 3.60 28.62 5.85

Reported Malaria 0.36 0.01 0.32 0.02 0.32 0.03

Workdays lost to Malaria 1.34 0.07 1.97 0.19 1.96 0.24

Expenditure on Malaria (Rs.) 105.18 9.31 110.49 10.49 151.79 39.08

observations 1867 799 283

 

 Figure 3. Conceptual framework to analyze the impact of mines on health for the population living in close proximity

The arrows indicate a causal link between different elements in the framework. A (-) sign indicates a negative relationship in the causal chain – for example, mining often involves clearing forests for excavation and thus proximity to mines reduces the availability of forest resources. On the other hand, an example of hypothesized positive relation between higher disposable income and higher adoption of prevention measures is indicated by a (+) sign.

Table 4. Probit model for incidence of ARI and malaria among individual family members (ivprobit in Stata)  

probit for ARI probit for malaria dy/dx st. error a p-val dy/dx st. error a p-val Dummy for male 0.034 0.057 0.533 0.054 0.054 0.320 Age 0.000 0.002 0.841 0.002 0.002 0.231 Scheduled Caste -0.116 0.179 0.516 -0.051 0.121 0.671 Scheduled Tribe -0.219 0.116 0.06 -0.178 0.093 0.056 Log of firewood consumed 0.009 0.033 0.783 Average irrigated land around village -0.034 0.008 0.000 Change in forest cover in a 2k buffer 0.004 0.005 0.353 Log of distance to road -0.057 0.054 0.296 -0.028 0.052 0.592 Log of distance to health center 0.033 0.046 0.474 0.017 0.037 0.643 Days worked in mine # -0.001 0.001 0.045 0.001 0.001 0.343 Distance to iron ore mine -0.023 0.008 0.006 0.019 0.007 0.003 Dummy for improved stove -0.757 0.475 0.101 Dummy for partition in kitchen -0.230 0.096 0.017 Dummy for use of mosquito nets -0.118 0.067 0.076 Number of observations 2949 2949 AIC 35241.6 35864.4 Log pseudo likelihood -17593.8 -17904.2

     # Days worked in mines instrumented in a 2SLS model. Instruments used are distance to iron ore mine, average number of hours worked in wage employment, dummy if land owned and years of education. a. Clustered robust standard errors by village

Table 5. 3SLS model for workdays lost due to ARI and malaria (reg3 in Stata)   Workdays lost to ARI Workdays lost to malaria coeff st err a p-val coeff st err a p-val Dummy for male -0.042 0.163 0.790 0.063 0.156 0.681 Age 0.007 0.004 0.110 0.005 0.004 0.238 Scheduled Caste -0.662 0.320 0.039 -0.310 0.308 0.305 Scheduled Tribe -0.794 0.174 0.000 -0.490 0.173 0.005 Log of firewood consumed -0.030 0.055 0.587 Average irrigated land around village -0.036 0.020 0.076 Change in forest cover in a 2k buffer 0.024 0.009 0.009 Log of distance to road -0.065 0.073 0.368 -0.018 0.072 0.802 Log of distance to health center 0.120 0.076 0.117 0.077 0.076 0.318 Days worked in mine -0.001 0.002 0.549 0.001 0.002 0.536 Distance to iron ore mine -0.026 0.014 0.066 -0.085 0.014 0.000 Dummy for improved stove -0.909 0.734 0.216 Dummy for partition in kitchen -0.289 0.150 0.054 Dummy for use of mosquito nets -0.254 0.151 0.090 constant 1.424 0.434 0.007 2.342 0.471 0.000 Chi-squared 35.34 0.03 79.83 Prob > Chi-Sq 0.000 0.000 Days worked in mine Distance to iron ore mine -1.441 0.188 0.000 Daily hours in wage employment 14.052 0.396 0.000 Dummy if land owned -7.007 2.332 0.003 Years of education 0.712 0.267 0.008 Constant 11.100 2.091 0.000 Chi-squared 1419.64 Prob > Chi-Sq 0.000 Number of observation 2949

a Clustered robust standard errors by village. # Days worked in mines instrumented in a 3SLS model. Instruments used are distance to nearest iron ore mine, average number of hours worked in wage employment, dummy if land owned and years of education. Adding more variables like Scheduled caste, Scheduled tribe, age and sex to the first stage equation of days worked in mines did not change the signs and significance of coefficients in the equation of workdays lost. Estimating the same models with the restricted samples where workdays lost due to ARI and malaria was <15 did not produce any significant changes to the results.

Table 6. Zero-inflated Negative Binomial model for workdays lost to ARI and Malaria  

Workdays lost to ARI Workdays lost to Malaria coeff st error b p-val coeff st error b p-val

Dummy for male -0.140 0.060 0.02 0.045 0.077 0.56 Age 0.001 0.003 0.76 0.001 0.002 0.73 Scheduled Caste -0.234 0.126 0.06 0.028 0.123 0.82 Scheduled Tribe -0.090 0.086 0.29 -0.124 0.147 0.40 Log of firewood consumed -0.055 0.031 0.07 Average % irrigated land in 2kbuffer 0.031 0.022 0.15 Decrease in forest cover 0.015 0.004 0.00 Log of distance to road 0.034 0.026 0.19 0.018 0.050 0.72 Log of distance to health center 0.102 0.049 0.04 0.045 0.053 0.39 Days worked in mine -0.004 0.007 0.50 0.000 0.001 0.84 Distance to iron ore mine -0.003 0.001 0.42 a -0.074 0.013 0.00 Dummy for improved stove -0.116 0.074 0.12 Dummy for partitioned kitchen -0.019 0.086 0.83 Dummy for use of mosquito nets -0.111 0.125 0.37 constant 1.846 0.241 0.00 1.815 0.389 0.00 Inflation equation: Household below poverty line 0.462 0.107 0.00 0.160 0.119 0.18 Log of distance to health center -0.064 0.087 0.46 0.097 0.050 0.05 Dummy for male 0.005 0.088 0.96 0.147 0.074 0.05 Age 0.007 0.005 0.12 -0.002 0.003 0.52 Education 0.035 0.016 0.03 -0.028 0.020 0.15 Scheduled Caste 0.161 0.347 0.64 0.506 0.244 0.04 Scheduled Tribe 0.482 0.195 0.01 0.218 0.209 0.30 Family size 0.195 0.039 0.00 0.073 0.037 0.05 constant 0.266 0.433 0.54 -0.318 0.324 0.33  a. Barely significant at the 15% level when the logarithm of distance to nearest iron ore mines is used. b. Clustered robust standard errors by village.    

   

Table 7. Probit models of adoption of preventive measures (ivprobit in Stata)  

Use of bed nets Use of improved stoves b Better kitchen construction dy/dx st erra p-val dy/dx st erra p-val dy/dx st erra p-val Dummy for land ownership 0.143 0.079 0.06 0.238 0.001 0.02 -0.002 0.051 0.95 Dummy for participation in micro-credit group c -0.123 0.074 0.09 0.013 0.093 0.88 Adoption by other members in same village 0.02 0.005 0.00 -0.113 0.304 0.44 0.007 0.006 0.24 Participate in social organizations 0.089 0.035 0.01 -0.082 0.081 0.15 0.002 0.041 0.96 Log of distance to markets -0.001 0.031 0.97 -0.278 0.154 0.59 -0.008 0.027 0.75 Number of rooms in the house 0.135 0.028 0.00 0.097 0.217 0.19 -0.171 0.019 0.00 Days worked in mines b 0.004 0.008 0.64 0.002 0.041 0.02 0.003 0.007 0.60 Number of observations 2949 528 2949 Wald Chi-Sq. 106.56 25.83 103.63 Log pseudolikelihood -1629.31 -5697.66 -1752.84 Wald test for exogeneity: Prob > Chi_sq 0.89 0.01 0.84

a. Clustered robust standard error by villages b. Days worked in mines is treated as endogenous and instruments used are – distance to iron ore mine and average amount spent by

household members in wage employment. c. None of the families that adopted improved stoves participated in micro-credit organizations. The variable drops out of the equation.